- Email: [email protected]
Moderators of delay tolerance in treatment-seeking cocaine users
Addictive Behaviors 32 (2007) 370 – 376
Short Communication
Moderators of delay tolerance in treatment-seeking cocaine users Ayesha Chawdhary a,b, Shelly L. Sayre a,b, Charles Green a,b, Joy M. Schmitz a,b, John Grabowski a,b, Marc E. Mooney c,* a
University of Texas Health Science Center–Houston Substance Abuse Research Center, USA b Department of Psychiatry and Behavioral Sciences, University of Texas, Houston, USA c Department of Psychiatry, University of Minnesota, Minneapolis, USA
Abstract A substantial amount of attrition in cocaine dependence treatment studies occurs between the initial telephone contact and the first evaluative clinic visit. While decreasing the wait to first visit can significantly reduce pre-intake attrition (PIA), little is known about other factors that moderate delay tolerance for first clinic visit. The current report uses data from 833 subjects who completed a first-contact telephone interview prior to an intake evaluation visit for cocaine use treatment research. Hierarchical logistic regression was used to assess three successive models to predict PIA, with the most inclusive model testing interactions between delay interval and seven predictors: age, gender, treatment motivation, recency of cocaine, alcohol, and tobacco use, and self-reported depression. Consistent with previous reports, greater delay to first clinic visit predicted PIA. However, no evidence for the moderating role of the selected factors was found. Overall, the utility of the logistic models, built on basic demographic and psychiatric factors, was poor, as evaluated using receiver–operator characteristic curves. Alternative factors must be examined to identify predictors that will increase probability of initial enrolment in cocaine-dependence clinical trials. D 2006 Elsevier Ltd. All rights reserved. Keywords: Attrition; Cocaine; Impulsivity; Prognostic modeling
* Corresponding author. Department of Psychiatry University of Minnesota, USA Tobacco Use Research Center 2701 University Avenue, S.E. Minneapolis, MN 55414. Tel.: +1 612 627 1822; fax: +1 612 627 4899. E-mail address: [email protected] (M.E. Mooney). 0306-4603/$ - see front matter D 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.addbeh.2006.03.044
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
371
1. Introduction High subject attrition poses a serious methodological and clinical problem in drug abuse treatment studies. In particular, problems of attrition in treatment-seeking cocaine users have been frequently described (Gainey, Wells, Hawkins, & Catalano, 1993; Pena et al., 1999). A consistent finding has been that the greater the delay between telephone screening and intake evaluation, the greater the chance of pre-intake attrition (PIA, Festinger, Lamb, Kountz, Kirby, & Marlowe, 1995; Siqueland et al., 1998; Woody, O’Hare, Mintz, & O’Brien, 1975). Conversely, accelerating treatment intake can significantly decrease PIA (Festinger, Lamb, Kirby, & Marlowe, 1996; Festinger, Lamb, Marlowe, & Kirby, 2002). Delineating additional factors and doing so more precisely would enhance recruitment resources and might ultimately facilitate treatment entry procedures across a range of venues. Cocaine users specifically represent a clinical population for whom failed treatment engagement may have dire consequences. To identify factors moderating delay tolerance for first clinic visit, we evaluated first visit attendance rates in 883 research treatment-seeking cocaine users. We assessed the moderating effects of multiple baseline variables on the effect of delay to first clinic visit on PIA.
2. Methods 2.1. Participants, procedures, and measures The University of Texas, Houston Committee for the Protection of Human Subjects approved the current research, conducted through Department of Psychiatry and Behavioral Sciences in the Texas Medical Center at Houston. Between February 1, 2004 and January 31, 2005, 1698 women and men responded to advertisements in the local media for three pharmacotherapy trials for cocaine dependence. Inclusion criteria for invitation to an initial clinic visit were: (a) use of cocaine in the last 30 days; (b) being between the ages of 18 and 60; (c) being willing to attend multiple weekly treatment sessions; and (d) being able to read and write in English. Exclusion criteria included: (a) evidence of abuse or dependence of other drugs in the preceding 30 days (except nicotine and alcohol; one protocol enrolled patients with problem levels of drinking); (b) active major mental illness including psychotic and affective disorders (one protocol enrolled patients with self-reported depression); and (c) significant untreated or uncontrolled medical problems. Clinic research assistants evaluated each caller using a specially designed computer-assisted telephone interviewing system (CATI, e.g., Choi, 2004), constructed in Microsoft Access 2000, for instant data entry and application of validation criteria and skip-out logic. The CATI questions included a range of content but for the purposes of the current report the following variables are considered: (a) delay period (i.e., days between telephone interview and scheduled appointment); (b) age; (c) sex; (d) use of cocaine, alcohol, and tobacco in the 7 days preceding telephone interview; (e) quantity/frequency of cocaine, alcohol and tobacco use (i.e., $/day, drinks/day, or cigarettes/day); and (f) treatment motivation (0 = no motivation reported; 1 = self-reported motivation for treatment, e.g., bI want to save my marriageQ). Race and ethnicity data were not collected. Subjects were scheduled according to the next available intake time, and not according to subject characteristics or requests. Attendance data were captured at the first clinic visit.
372
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
2.2. Analyses All analyses were conducted using the Statistical Analysis System, Version 9.1 (SAS Institute Inc., 2004). Values of p b .05 were considered statistically significant, based on two-tailed tests. Incremental model fit of logistic regressions was assessed using the differences between log-likelihood statistics and chi-square statistics. Effect sizes were expressed as odds ratios with 95% confidence intervals. The utility of each regression model was also evaluated in terms of area under-the-curve (AUC) and presented using receiver–operator characteristic (ROC) curves. Performance of a prognostic model, in terms of AUC, can be evaluated as follows: N .90 = bhighQ accuracy; .90 to .70 = busefulQ accuracy; .69 to .51. = blowQ accuracy; and .50 reflects a diagnostic test with chance accuracy (Swets, 1988).
3. Results 3.1. Participant characteristics Of a total number of initial telephone interviews conducted, (N = 1437), 883 callers completed the interview and received an initial clinic visit appointment. The remaining 265 callers either terminated the telephone interview or did not accept a clinic visit appointment, and thus were coded ineligible for this study. Average age of eligible subjects was 38.1 years (SD = 8.4); most were male (65.8%), see Table 1. Most had used cocaine in the preceding week (90.4%) as well as alcohol (65.2%) and tobacco (50.1%). Few participants reported a motivating reason for seeking treatment, and most callers did not report current depression. Table 1 Participant characteristics Telephone screena Scheduled wait (days) Age (years) Sex (% female) Use in last 7 days % Cocaine % Alcohol % Tobacco Quantity/frequencyb Cocaine ($/day)c Alcohol (drinks/day) Tobacco (cigarettes/day) % Motivation % Self-reported depression
Initial clinic visit
Ineligible n = 554
Eligible n = 883
Absent n = 516
Present n = 367
– 37.8 (9.6) 26.1
6.2 (4.6) 38.1 (8.4) 34.2
6.7 (4.7) 37.2 (8.5) 29.3
5.5 (4.5) 39.3 (8.0) 21.6
72.2 52.5 39.7
90.4 65.2 50.1
88.0 68.2 51.2
93.7 61.0 48.5
40 (0.2, 2,000) 6.0 (1.1, 27.4) 20.0 (2.0, 70.0) – 18.6
43 (1.0, 3,000) 4.6 (1.1, 28.6) 20.0 (1.1, 90.0) 28.8 10.3
46 (1.0, 3,000) 4.8 (1.1, 27.4) 20.0 (1.1, 90.0) 29.9 13.2
41(1.3, 1,400) 4.4 (1.1, 28.6) 20.0 (1.7, 80.0) 31.1 6.3
Note. A total of 1433 subjects completed screening, of whom 833 were eligible and scheduled a first clinic visit. a265 subjects who declined participation, who were eligible but who never scheduled a first visit, or who hung up are not included. bDue to uncertainty about the equivalence of units, quantity/frequency data were not employed as predictors in regression models. c Cocaine quantity/frequency is only reported for those describing use in terms of US$ and not non-standard units (e.g., b8 ballsQ, rocks, grams).
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
373
3.2. Prognostic modeling 3.2.1. Model comparisons In order to identify predictors of PIA, three logistic models were evaluated in hierarchical fashion beginning only with the delay interval as a predictor and culminating in a multifactorial model (see Table 2 for test statistics, p-values, and odds ratio effect sizes). In comparison to the unconditional or intercept only model, the addition of delay as a predictor substantially improved the fit of the model (Dv 2(1) = 18.102, p b .0001) with this single-predictor model demonstrating adequate fit (Hosmer–Lemshow v 2(8) = 13.328, n.s.). For each increment in delay to first visit, an associated increase in odds of PIA occurred. The addition of a block of seven predictors including age, sex, motivation, past-week cocaine use, past-week alcohol use, past-week cigarette use, and self-reported depression reliably improved the model fit, (Dv 2(8) = 61.7, p b .0001). The general fit of the model to the data remained adequate (Hosmer–Lemshow v 2(8) = 11.1, n.s.). Greater delay to first visit continued to be positively associated with PIA. Use of cocaine in the 7 days prior to interview was negatively associated with PIA, while use of alcohol in this period was positively associated with PIA. Decreasing age was associated with greater likelihood of PIA. Self-reported depression greater than doubled the likelihood of PIA. Finally, the moderating effects of the seven predictors were evaluated by including interaction terms with delay interval. However, none of the interaction terms were significant, and allowing for moderator terms did not improve the prognostic model of PIA, (Dv 2(14) = 5.95, n.s.). 3.2.2. Model utility The utility of the logistic regression models was examined in terms of correct classification of subjects (i.e., absence from first visit), shown with receiver–operator characteristic curves (see Fig. 1). Table 2 Prognostic logistic regression models of pre-intake attrition Model 1 Wait Age Sex Motivation Cocaine (7 days) Alcohol (7 days) Cigarettes (7 days) Depression Age delay Sex delay Motivation delay Cocaine delay Alcohol delay Cigarettes delay Depression delay
Model 2
Model 3
OR, 95% C.I.
OR, 95% C.I.
1.07 (1.03, 1.10)y – – – – – _ _ – – – – – _ _
1.06 0.97 1.14 1.05 0.48 1.59 1.07 2.09 – – – – – _ _
(1.02, (0.95, (0.97, (0.90, (0.28, (1.16, (0.81, (1.23,
1.09)y 0.99)y 1.35) 1.23) 0.82)** 2.17) ** 1.43) 3.54)**
OR, 95% C.I. 1.19 0.98 1.13 1.14 0.38 1.68 1.54 1.57 1.00 1.00 0.98 1.03 0.99 0.94 1.04
(0.97, (0.95, (0.84, (0.87, (0.16, (0.99, (0.94, (0.61, (0.99, (0.96, (0.95, (0.92, (0.92, (0.88, (0.91,
1.46) 1.01) 1.52) 1.49) 0.90)* 2.82)* 2.51) 4.04) 1.00) 1.04) 1.02) 1.15) 1.08) 1.01) 1.19)
Note. *p b .05. **p b .01. yp b .001. All chi-square tests on 1 degree of freedom. Three logistic models were evaluated to predict pre-intake attrition (PIA) in a hierarchical fashion beginning only with delay interval as a predictor (i.e., Model 1) and culminating in a multifactorial model (i.e., Model 3).
374
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
Fig. 1. Receiver–operator characteristic curves for three logistic regression models (see Table 2). A diagnostic test with chance accuracy is represented by the 45-degree diagonal line, i.e., AUC = .5. All three models had blowQ diagnostic performance (see text and Swets, 1988).
AUC quantities for Models 1 (AUC = .597), 2 (AUC = .645), and 3 (AUC = .648) were all blow.Q Taken together these indices suggest that while there is some reliable prediction on the part of some of the variables in the logistic regressions, they still model only a small amount of the variability in PIA.
4. Discussion 4.1. Delay tolerance Using telephone screening data collected over a one-year period, we confirmed previous findings that longer delays to first clinic visit were associated with PIA in treatment-seeking cocaine users. In the light of increasing findings concerning the connection between substance abuse and impulsivity (Moeller, Barratt, Dougherty, Schmitz, & Swann, 2001), it is not surprising to observe an association between delay to first visit and attrition. Most conceptualizations of impulsivity postulate greater likelihood in substance use disorders. An important element of impulsivity is a preference for more immediate rewards and a tendency to discount more temporally distant rewards (Moeller et al., 2001). To the extent that btreatmentQ is viewed as potentially rewarding, impulsivity conceptualizations may be applicable. While in the current sample, the magnitude of the effect of delay was relatively small, expeditious first visit scheduling is likely a palliative, but by no means a sufficient remedy for high rates of PIA. 4.2. Other predictors Other predictors were also evaluated that bore reliable but mostly small relationships to PIA, with the exception of self-reported depression. The relationship between depressed mood and cocaine use has been extensively described (Rounsaville, 2004). Those reporting feeling depressed were twice as prone to attrition prior to the initial clinic visit. More detailed assessment of mood and negative affect may be
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
375
another approach to identifying those in need of more immediate evaluation. Younger age was also associated with greater likelihood of PIA, which may indirectly be supportive of reports that older subjects are more likely to remain in treatment. 4.3. Moderators of delay tolerance The effect of delay to first clinic visit on PIA was not influenced by any of the individual subject characteristics explored in this study. The failure to identify moderators of delay in treatment is somewhat surprising, given our ostensibly large sample size. While considered useful to identify patients for whom the delay factor more strongly affects attendance, the present findings do not point to clear matching hypotheses. It seems more likely that, while one or two factors may generally account for a larger portion of the variance, the myriad factors that may contribute make the task of clear identification difficult. Moving beyond demographic and substance use factors as predictors may prove to be more clinically useful and relevant. 4.4. Limitations Several limitations to the current paper require discussion. All predictors used in prognostic models were obtained via self-report, and may be invalid due to deception, mnemonic bias, or memory disruption arising from substance abuse. Second, the findings of prognostic factors were not replicated in a reserve sample. Third, race and ethnicity data were not collected in this study. This important variable should be included in future studies to help understand its role in health disparities. 4.5. Conclusions The likelihood of finding single factors that strongly predict failure to attend an initial clinic visit is low, suggesting the need for multivariate models. However, the models evaluated in this report performed at essentially chance levels, despite the identification of multiple statistically significant predictors. Other measurement targets including improved measures of motivation and mood as well as practical barriers to attendance, including transportation and childcare, may improve the forecasting of pre-intake attrition. Acknowledgements This research was supported by the NIDA grants DA-13333, DA-009262, DA-16305, DA-15433, DA-08654, DA-11216. We thank the participants for taking part in these studies. References Choi, B. C. (2004). Computer assisted telephone interviewing (CATI) for health surveys in public health surveillance: Methodological issues and challenges ahead. Chronic Diseases in Canada, 25(2), 21 – 27. Festinger, D. S., Lamb, R. J., Kirby, K. C., & Marlowe, D. B. (1996). The accelerated intake: A method for increasing initial attendance to outpatient cocaine treatment. Journal of Applied Behavior Analysis, 29(3), 387 – 389.
376
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
Festinger, D. S., Lamb, R. J., Kountz, M. R., Kirby, K. C., & Marlowe, D. (1995). Pretreatment dropout as a function of treatment delay and client variables. Addictive Behaviors, 20(1), 111 – 115. Festinger, D. S., Lamb, R. J., Marlowe, D. B., & Kirby, K. C. (2002). From telephone to office — intake attendance as a function of appointment delay. Addictive Behaviors, 27(1), 131 – 137. Gainey, R. R., Wells, E. A., Hawkins, J. D., & Catalano, R. F. (1993). Predicting treatment retention among cocaine users. International Journal of the Addictions, 28(6), 487 – 505. Moeller, F. G., Barratt, E. S., Dougherty, D. M., Schmitz, J. M., & Swann, A. C. (2001). Psychiatric aspects of impulsivity. American Journal of Psychiatry, 158(11), 1783 – 1793. Pena, J. M., Franklin, R. R., Rice, J. C., Foulks, E. F., Bland, I. J., Shervington, D., et al. (1999). A two-rate hypothesis for patterns of retention and psychosocial treatments of cocaine dependence: Findings from a study of African-American men and a review of the published data. American Journal on Addictions, 8(4), 319 – 331. Rounsaville, B. J. (2004). Treatment of cocaine dependence and depression. Biological Psychiatry, 56(10), 803 -- 809. SAS Institute Inc. (2004). The SAS system for windows (Version 9.1). Cary, NC7 SAS Institute, Inc. Siqueland, L., Crits-Christoph, P., Frank, A., Daley, D., Weiss, R., Chittams, J., et al. (1998). Predictors of dropout from psychosocial treatment of cocaine dependence. Drug and Alcohol Dependence, 52(1), 1 – 13. Swets, J. A. (1988). Measuring the accuracy of diagnostic systems. Science, 240(4857), 1285 – 1293. Woody, G., O’Hare, K., Mintz, J., & O’Brien, C. (1975). Rapid intake: A method for increasing retention rate of heroin addicts seeking methadone treatment. Comprehensive Psychiatry, 16(2), 165 – 169.
Short Communication
Moderators of delay tolerance in treatment-seeking cocaine users Ayesha Chawdhary a,b, Shelly L. Sayre a,b, Charles Green a,b, Joy M. Schmitz a,b, John Grabowski a,b, Marc E. Mooney c,* a
University of Texas Health Science Center–Houston Substance Abuse Research Center, USA b Department of Psychiatry and Behavioral Sciences, University of Texas, Houston, USA c Department of Psychiatry, University of Minnesota, Minneapolis, USA
Abstract A substantial amount of attrition in cocaine dependence treatment studies occurs between the initial telephone contact and the first evaluative clinic visit. While decreasing the wait to first visit can significantly reduce pre-intake attrition (PIA), little is known about other factors that moderate delay tolerance for first clinic visit. The current report uses data from 833 subjects who completed a first-contact telephone interview prior to an intake evaluation visit for cocaine use treatment research. Hierarchical logistic regression was used to assess three successive models to predict PIA, with the most inclusive model testing interactions between delay interval and seven predictors: age, gender, treatment motivation, recency of cocaine, alcohol, and tobacco use, and self-reported depression. Consistent with previous reports, greater delay to first clinic visit predicted PIA. However, no evidence for the moderating role of the selected factors was found. Overall, the utility of the logistic models, built on basic demographic and psychiatric factors, was poor, as evaluated using receiver–operator characteristic curves. Alternative factors must be examined to identify predictors that will increase probability of initial enrolment in cocaine-dependence clinical trials. D 2006 Elsevier Ltd. All rights reserved. Keywords: Attrition; Cocaine; Impulsivity; Prognostic modeling
* Corresponding author. Department of Psychiatry University of Minnesota, USA Tobacco Use Research Center 2701 University Avenue, S.E. Minneapolis, MN 55414. Tel.: +1 612 627 1822; fax: +1 612 627 4899. E-mail address: [email protected] (M.E. Mooney). 0306-4603/$ - see front matter D 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.addbeh.2006.03.044
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
371
1. Introduction High subject attrition poses a serious methodological and clinical problem in drug abuse treatment studies. In particular, problems of attrition in treatment-seeking cocaine users have been frequently described (Gainey, Wells, Hawkins, & Catalano, 1993; Pena et al., 1999). A consistent finding has been that the greater the delay between telephone screening and intake evaluation, the greater the chance of pre-intake attrition (PIA, Festinger, Lamb, Kountz, Kirby, & Marlowe, 1995; Siqueland et al., 1998; Woody, O’Hare, Mintz, & O’Brien, 1975). Conversely, accelerating treatment intake can significantly decrease PIA (Festinger, Lamb, Kirby, & Marlowe, 1996; Festinger, Lamb, Marlowe, & Kirby, 2002). Delineating additional factors and doing so more precisely would enhance recruitment resources and might ultimately facilitate treatment entry procedures across a range of venues. Cocaine users specifically represent a clinical population for whom failed treatment engagement may have dire consequences. To identify factors moderating delay tolerance for first clinic visit, we evaluated first visit attendance rates in 883 research treatment-seeking cocaine users. We assessed the moderating effects of multiple baseline variables on the effect of delay to first clinic visit on PIA.
2. Methods 2.1. Participants, procedures, and measures The University of Texas, Houston Committee for the Protection of Human Subjects approved the current research, conducted through Department of Psychiatry and Behavioral Sciences in the Texas Medical Center at Houston. Between February 1, 2004 and January 31, 2005, 1698 women and men responded to advertisements in the local media for three pharmacotherapy trials for cocaine dependence. Inclusion criteria for invitation to an initial clinic visit were: (a) use of cocaine in the last 30 days; (b) being between the ages of 18 and 60; (c) being willing to attend multiple weekly treatment sessions; and (d) being able to read and write in English. Exclusion criteria included: (a) evidence of abuse or dependence of other drugs in the preceding 30 days (except nicotine and alcohol; one protocol enrolled patients with problem levels of drinking); (b) active major mental illness including psychotic and affective disorders (one protocol enrolled patients with self-reported depression); and (c) significant untreated or uncontrolled medical problems. Clinic research assistants evaluated each caller using a specially designed computer-assisted telephone interviewing system (CATI, e.g., Choi, 2004), constructed in Microsoft Access 2000, for instant data entry and application of validation criteria and skip-out logic. The CATI questions included a range of content but for the purposes of the current report the following variables are considered: (a) delay period (i.e., days between telephone interview and scheduled appointment); (b) age; (c) sex; (d) use of cocaine, alcohol, and tobacco in the 7 days preceding telephone interview; (e) quantity/frequency of cocaine, alcohol and tobacco use (i.e., $/day, drinks/day, or cigarettes/day); and (f) treatment motivation (0 = no motivation reported; 1 = self-reported motivation for treatment, e.g., bI want to save my marriageQ). Race and ethnicity data were not collected. Subjects were scheduled according to the next available intake time, and not according to subject characteristics or requests. Attendance data were captured at the first clinic visit.
372
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
2.2. Analyses All analyses were conducted using the Statistical Analysis System, Version 9.1 (SAS Institute Inc., 2004). Values of p b .05 were considered statistically significant, based on two-tailed tests. Incremental model fit of logistic regressions was assessed using the differences between log-likelihood statistics and chi-square statistics. Effect sizes were expressed as odds ratios with 95% confidence intervals. The utility of each regression model was also evaluated in terms of area under-the-curve (AUC) and presented using receiver–operator characteristic (ROC) curves. Performance of a prognostic model, in terms of AUC, can be evaluated as follows: N .90 = bhighQ accuracy; .90 to .70 = busefulQ accuracy; .69 to .51. = blowQ accuracy; and .50 reflects a diagnostic test with chance accuracy (Swets, 1988).
3. Results 3.1. Participant characteristics Of a total number of initial telephone interviews conducted, (N = 1437), 883 callers completed the interview and received an initial clinic visit appointment. The remaining 265 callers either terminated the telephone interview or did not accept a clinic visit appointment, and thus were coded ineligible for this study. Average age of eligible subjects was 38.1 years (SD = 8.4); most were male (65.8%), see Table 1. Most had used cocaine in the preceding week (90.4%) as well as alcohol (65.2%) and tobacco (50.1%). Few participants reported a motivating reason for seeking treatment, and most callers did not report current depression. Table 1 Participant characteristics Telephone screena Scheduled wait (days) Age (years) Sex (% female) Use in last 7 days % Cocaine % Alcohol % Tobacco Quantity/frequencyb Cocaine ($/day)c Alcohol (drinks/day) Tobacco (cigarettes/day) % Motivation % Self-reported depression
Initial clinic visit
Ineligible n = 554
Eligible n = 883
Absent n = 516
Present n = 367
– 37.8 (9.6) 26.1
6.2 (4.6) 38.1 (8.4) 34.2
6.7 (4.7) 37.2 (8.5) 29.3
5.5 (4.5) 39.3 (8.0) 21.6
72.2 52.5 39.7
90.4 65.2 50.1
88.0 68.2 51.2
93.7 61.0 48.5
40 (0.2, 2,000) 6.0 (1.1, 27.4) 20.0 (2.0, 70.0) – 18.6
43 (1.0, 3,000) 4.6 (1.1, 28.6) 20.0 (1.1, 90.0) 28.8 10.3
46 (1.0, 3,000) 4.8 (1.1, 27.4) 20.0 (1.1, 90.0) 29.9 13.2
41(1.3, 1,400) 4.4 (1.1, 28.6) 20.0 (1.7, 80.0) 31.1 6.3
Note. A total of 1433 subjects completed screening, of whom 833 were eligible and scheduled a first clinic visit. a265 subjects who declined participation, who were eligible but who never scheduled a first visit, or who hung up are not included. bDue to uncertainty about the equivalence of units, quantity/frequency data were not employed as predictors in regression models. c Cocaine quantity/frequency is only reported for those describing use in terms of US$ and not non-standard units (e.g., b8 ballsQ, rocks, grams).
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
373
3.2. Prognostic modeling 3.2.1. Model comparisons In order to identify predictors of PIA, three logistic models were evaluated in hierarchical fashion beginning only with the delay interval as a predictor and culminating in a multifactorial model (see Table 2 for test statistics, p-values, and odds ratio effect sizes). In comparison to the unconditional or intercept only model, the addition of delay as a predictor substantially improved the fit of the model (Dv 2(1) = 18.102, p b .0001) with this single-predictor model demonstrating adequate fit (Hosmer–Lemshow v 2(8) = 13.328, n.s.). For each increment in delay to first visit, an associated increase in odds of PIA occurred. The addition of a block of seven predictors including age, sex, motivation, past-week cocaine use, past-week alcohol use, past-week cigarette use, and self-reported depression reliably improved the model fit, (Dv 2(8) = 61.7, p b .0001). The general fit of the model to the data remained adequate (Hosmer–Lemshow v 2(8) = 11.1, n.s.). Greater delay to first visit continued to be positively associated with PIA. Use of cocaine in the 7 days prior to interview was negatively associated with PIA, while use of alcohol in this period was positively associated with PIA. Decreasing age was associated with greater likelihood of PIA. Self-reported depression greater than doubled the likelihood of PIA. Finally, the moderating effects of the seven predictors were evaluated by including interaction terms with delay interval. However, none of the interaction terms were significant, and allowing for moderator terms did not improve the prognostic model of PIA, (Dv 2(14) = 5.95, n.s.). 3.2.2. Model utility The utility of the logistic regression models was examined in terms of correct classification of subjects (i.e., absence from first visit), shown with receiver–operator characteristic curves (see Fig. 1). Table 2 Prognostic logistic regression models of pre-intake attrition Model 1 Wait Age Sex Motivation Cocaine (7 days) Alcohol (7 days) Cigarettes (7 days) Depression Age delay Sex delay Motivation delay Cocaine delay Alcohol delay Cigarettes delay Depression delay
Model 2
Model 3
OR, 95% C.I.
OR, 95% C.I.
1.07 (1.03, 1.10)y – – – – – _ _ – – – – – _ _
1.06 0.97 1.14 1.05 0.48 1.59 1.07 2.09 – – – – – _ _
(1.02, (0.95, (0.97, (0.90, (0.28, (1.16, (0.81, (1.23,
1.09)y 0.99)y 1.35) 1.23) 0.82)** 2.17) ** 1.43) 3.54)**
OR, 95% C.I. 1.19 0.98 1.13 1.14 0.38 1.68 1.54 1.57 1.00 1.00 0.98 1.03 0.99 0.94 1.04
(0.97, (0.95, (0.84, (0.87, (0.16, (0.99, (0.94, (0.61, (0.99, (0.96, (0.95, (0.92, (0.92, (0.88, (0.91,
1.46) 1.01) 1.52) 1.49) 0.90)* 2.82)* 2.51) 4.04) 1.00) 1.04) 1.02) 1.15) 1.08) 1.01) 1.19)
Note. *p b .05. **p b .01. yp b .001. All chi-square tests on 1 degree of freedom. Three logistic models were evaluated to predict pre-intake attrition (PIA) in a hierarchical fashion beginning only with delay interval as a predictor (i.e., Model 1) and culminating in a multifactorial model (i.e., Model 3).
374
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
Fig. 1. Receiver–operator characteristic curves for three logistic regression models (see Table 2). A diagnostic test with chance accuracy is represented by the 45-degree diagonal line, i.e., AUC = .5. All three models had blowQ diagnostic performance (see text and Swets, 1988).
AUC quantities for Models 1 (AUC = .597), 2 (AUC = .645), and 3 (AUC = .648) were all blow.Q Taken together these indices suggest that while there is some reliable prediction on the part of some of the variables in the logistic regressions, they still model only a small amount of the variability in PIA.
4. Discussion 4.1. Delay tolerance Using telephone screening data collected over a one-year period, we confirmed previous findings that longer delays to first clinic visit were associated with PIA in treatment-seeking cocaine users. In the light of increasing findings concerning the connection between substance abuse and impulsivity (Moeller, Barratt, Dougherty, Schmitz, & Swann, 2001), it is not surprising to observe an association between delay to first visit and attrition. Most conceptualizations of impulsivity postulate greater likelihood in substance use disorders. An important element of impulsivity is a preference for more immediate rewards and a tendency to discount more temporally distant rewards (Moeller et al., 2001). To the extent that btreatmentQ is viewed as potentially rewarding, impulsivity conceptualizations may be applicable. While in the current sample, the magnitude of the effect of delay was relatively small, expeditious first visit scheduling is likely a palliative, but by no means a sufficient remedy for high rates of PIA. 4.2. Other predictors Other predictors were also evaluated that bore reliable but mostly small relationships to PIA, with the exception of self-reported depression. The relationship between depressed mood and cocaine use has been extensively described (Rounsaville, 2004). Those reporting feeling depressed were twice as prone to attrition prior to the initial clinic visit. More detailed assessment of mood and negative affect may be
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
375
another approach to identifying those in need of more immediate evaluation. Younger age was also associated with greater likelihood of PIA, which may indirectly be supportive of reports that older subjects are more likely to remain in treatment. 4.3. Moderators of delay tolerance The effect of delay to first clinic visit on PIA was not influenced by any of the individual subject characteristics explored in this study. The failure to identify moderators of delay in treatment is somewhat surprising, given our ostensibly large sample size. While considered useful to identify patients for whom the delay factor more strongly affects attendance, the present findings do not point to clear matching hypotheses. It seems more likely that, while one or two factors may generally account for a larger portion of the variance, the myriad factors that may contribute make the task of clear identification difficult. Moving beyond demographic and substance use factors as predictors may prove to be more clinically useful and relevant. 4.4. Limitations Several limitations to the current paper require discussion. All predictors used in prognostic models were obtained via self-report, and may be invalid due to deception, mnemonic bias, or memory disruption arising from substance abuse. Second, the findings of prognostic factors were not replicated in a reserve sample. Third, race and ethnicity data were not collected in this study. This important variable should be included in future studies to help understand its role in health disparities. 4.5. Conclusions The likelihood of finding single factors that strongly predict failure to attend an initial clinic visit is low, suggesting the need for multivariate models. However, the models evaluated in this report performed at essentially chance levels, despite the identification of multiple statistically significant predictors. Other measurement targets including improved measures of motivation and mood as well as practical barriers to attendance, including transportation and childcare, may improve the forecasting of pre-intake attrition. Acknowledgements This research was supported by the NIDA grants DA-13333, DA-009262, DA-16305, DA-15433, DA-08654, DA-11216. We thank the participants for taking part in these studies. References Choi, B. C. (2004). Computer assisted telephone interviewing (CATI) for health surveys in public health surveillance: Methodological issues and challenges ahead. Chronic Diseases in Canada, 25(2), 21 – 27. Festinger, D. S., Lamb, R. J., Kirby, K. C., & Marlowe, D. B. (1996). The accelerated intake: A method for increasing initial attendance to outpatient cocaine treatment. Journal of Applied Behavior Analysis, 29(3), 387 – 389.
376
A. Chawdhary et al. / Addictive Behaviors 32 (2007) 370–376
Festinger, D. S., Lamb, R. J., Kountz, M. R., Kirby, K. C., & Marlowe, D. (1995). Pretreatment dropout as a function of treatment delay and client variables. Addictive Behaviors, 20(1), 111 – 115. Festinger, D. S., Lamb, R. J., Marlowe, D. B., & Kirby, K. C. (2002). From telephone to office — intake attendance as a function of appointment delay. Addictive Behaviors, 27(1), 131 – 137. Gainey, R. R., Wells, E. A., Hawkins, J. D., & Catalano, R. F. (1993). Predicting treatment retention among cocaine users. International Journal of the Addictions, 28(6), 487 – 505. Moeller, F. G., Barratt, E. S., Dougherty, D. M., Schmitz, J. M., & Swann, A. C. (2001). Psychiatric aspects of impulsivity. American Journal of Psychiatry, 158(11), 1783 – 1793. Pena, J. M., Franklin, R. R., Rice, J. C., Foulks, E. F., Bland, I. J., Shervington, D., et al. (1999). A two-rate hypothesis for patterns of retention and psychosocial treatments of cocaine dependence: Findings from a study of African-American men and a review of the published data. American Journal on Addictions, 8(4), 319 – 331. Rounsaville, B. J. (2004). Treatment of cocaine dependence and depression. Biological Psychiatry, 56(10), 803 -- 809. SAS Institute Inc. (2004). The SAS system for windows (Version 9.1). Cary, NC7 SAS Institute, Inc. Siqueland, L., Crits-Christoph, P., Frank, A., Daley, D., Weiss, R., Chittams, J., et al. (1998). Predictors of dropout from psychosocial treatment of cocaine dependence. Drug and Alcohol Dependence, 52(1), 1 – 13. Swets, J. A. (1988). Measuring the accuracy of diagnostic systems. Science, 240(4857), 1285 – 1293. Woody, G., O’Hare, K., Mintz, J., & O’Brien, C. (1975). Rapid intake: A method for increasing retention rate of heroin addicts seeking methadone treatment. Comprehensive Psychiatry, 16(2), 165 – 169.